The common pattern across all of these seems to be filesystem and network ACLs enforced by the OS, not a separate kernel or hardware boundary. A determined attacker who already has code execution on your machine could potentially bypass Seatbelt or Landlock restrictions through privilege escalation. But that is not the threat model. The threat is an AI agent that is mostly helpful but occasionally careless or confused, and you want guardrails that catch the common failure modes - reading credentials it should not see, making network calls it should not make, writing to paths outside the project.
export OPENCLAW_STATE_DIR="$PWD/.openclaw_data"
，推荐阅读heLLoword翻译官方下载获取更多信息
Feb 25, 2026 6:00 AM
Update, February 27, 9PM ET: This story was updated twice after publish. First at 6PM ET to include a link to and quotes from Hegseth about the designation of Anthropic as a supply chain risk. Later, a quote from Anthropic was added, along with a link to the company’s blog post on the subject.